THE FINITE DIFFERENCE CONTRAST SOURCE INVERSION WITH SUPER MEMORY HYBRID CONJUGATE GRADIENT METHOD

摘要

The finite difference contrast source inversion (FDCSI) is an algorithm to solve the wave equation inverse scattering problem. This algorithm's forward operator is only related to the background medium, which does not change during the iterative optimization process. Therefore, an LU decomposition is required only once for the forward operator, which has lower computation cost. Because of finite difference operator, FDCSI can be applied to inhomogeneous background medium. FDCSI transforms the inverse scattering problem of wave equation into an optimization problem, which can be solved by conjugate gradient method. But conventional conjugate gradient method converges slowly, which affects computing efficiency, and the Newton method increases computation and memory. In order to improve the convergence speed for frequency domain acoustic equation, the super memory hybrid conjugate gradient method (SMHCG) is introduced into FDCSI. SMHCG is improved on the basis of the super memory gradient method to adapt to FDCSI. SMHCG accelerates the convergence of objective function without any increase computation and memory. The advantages of SMHCG had been verified on the Marmousi model.